adtestbench Higgsfield
EN

AI news · Google · Project Suncatcher

Google’s AI satellite sends its first answer from orbit

A Gemma 4 model running on a Google TPU in space wrote a haiku, an early milestone in Google’s research into running AI computing in orbit.

By Alexandre S. , 02:30 UTC

Google says the prototype satellite for Project Suncatcher, its research effort to run AI computing in space, has answered its first prompt. In a blog post on 30 October 2026, the company said engineers uplinked a request to the Gemma 4 open model running on the satellite’s TPU, Google’s AI chip, and received a haiku back as it passed over Svalbard, Norway. Google calls it the first time a TPU of the kind used in its data centres has answered an AI prompt in space.

Retro-futurist illustration: a striped sunset over a grid horizon under a starry sky, with a satellite standing on the horizon.
Drawn by adtestbench from “Google receives response from Project Suncatcher prototype satellite in space”,

Two linked satellites are planned for 2027, once trials show how the chips cope with heat and radiation, The Verge reports. Since the 1 October launch, the team has switched on the satellite’s host computer and TPU trays and run early machine learning jobs, according to Google.

The economics remain far off. Google’s own sums say the launch cost of computing in orbit could be comparable to a ground data centre’s electricity spending once launches fall to about $200 a kilogram, which it expects around 2035, The Register notes, against a retail launch price today of roughly $7,000 a kilogram, by The Register’s estimate.

Google’s paid AI still runs on the ground. Our 31 October check lists Veo 3.1 Fast at $0.10 a second of 720p video on the Gemini API and Veo 3.1 Lite at $0.05.

With Google placing cost parity for orbital computing around 2035 and its next step two linked satellites in 2027, the Veo prices on our sheet will be set by ground data centres for years yet.